🎉 🎉 RAPIDMINER 9.10 IS OUT!!! 🎉🎉

Download the latest version helping analytics teams accelerate time-to-value for streaming and IIOT use cases.

CLICK HERE TO DOWNLOAD

"What are the calculations of a neural network output?"

tigerwoodstigerwoods Member Posts: 3 Contributor I
edited June 2019 in Help
Hi,

I'm using a neural network to predict a continuous variable. My model is the following:

Node 1 (Sigmoid)
----------------
A -0,311
B -0,183
C -0,391
D -1,448
Bias -2,001

Node 2 (Sigmoid)
----------------
A 0,124
B -0,428
C -0,291
D -3,069
Bias -2,814

Node 3 (Sigmoid)
----------------
A -0,924
B -1,104
C -0,006
D -0,421
Bias -1,557


Output
======

Regression (Linear)
-------------------
Node 1 -0,583
Node 2 -2,161
Node 3 0,88
Threshold 0,533

With this model I want to predict new values, imagine for
A -1,834944684
B -1,754940513
C -0,312412486
D -1,275034298

Rapidminer gives me the result: 15,47%, but I'm not abble to arrive to this solution by the calculations.

My calculations are:

Value of Node 1 = 1/(1+exp(-(bias+sumproduct(coeficients in node 1; new values for A B C D)))
And I proceed similary for the other 2 nodes.

Value of output:
0,533 + (-0,583 )* value of node1 + (-2,161)*value of node2 + 0,88*value of node3
My result is: -88,8%

Can you help me to find my error in calculations?

Thank you very much.

Rufo

Answers

  • MariusHelfMariusHelf RapidMiner Certified Expert, Member Posts: 1,869   Unicorn
    Hi Rufo,

    maybe this is caused by a bug in the Neural Net operator that has already been fixed and will be included in the next release. Does the Neural Net work as expected if you apply it on the same data on which you trained it? If yes, then your problem will be fixed with the next release. Otherwise we will have to investigate further.

    Best regards,
    Marius
Sign In or Register to comment.